# Registered second-price-source volatility-sizing check v1

## Hypothesis and fixed definitions

Challenge the promising 60% budget / 60-day volatility sizing result by evaluating all ten unchanged models from VOLATILITY_SIZING_PROTOCOL.md on a second exchange's price history. Preserve all six volatility candidates and four fixed/hold controls, including failed lookbacks. Import the frozen signals, account, metrics and assumptions without modifying those helpers. No parameter selection, new targets, altered cadence or optimization. Historical periods are already explored; a second source is a robustness check, not an untouched independent market outcome.

## Frozen data intake

Use the 116 raw Binance BTCUSDT 4H archives already downloaded and verified in data/spot-flow-20260908-v1.1/. Require that receipt's SHA256 043f633815be40692c8deb163ba48fd19f2bd56a6898f296804df68d8869c60e and normalized flow SHA256 fd53b3fd0b7c3e14638f1d39c8e990114105228a62f5fffe1f2e7809e1ca9dd8 match before reading prices. Validate every raw archive against both the receipt hash and its published CHECKSUM; retain the raw-source paths and hashes in the derived receipt.

Reuse the exact existing schema/OHLC/unit/ordering validation and data-only quarantine rules. Extract OHLC only for the same 19,817 accepted timestamps as the normalized flow. The same nineteen incomplete/off-boundary bars remain excluded, with the same recorded quarantine details and 36 missing slots. Do not round timestamps, impute, smooth, clean price outliers based on returns, or change either source. Pre-2025 archive timestamps are milliseconds; 2025 onward are microseconds. Preserve new derived OHLC in a separately frozen directory and commit its identity before calculating strategy outcomes.

BTCUSD reference prices remain the two separate frozen input sets used in the first sizing study: canonical historical rows ending April 28, 2026 and recent market-cache prices with full warmup for the June 11-September 8 replay. Verify all previous source/input/provenance hashes. Original price repairs stay unchanged. Binance uses its own full accepted history from August 2017 for feature warmup, including for the separate recent window.

## Calendar windows and missing bars

Use the fourteen windows returned by evaluate_spot_flow.candidate_windows: 2018 onward, 2019-2022, 2023 onward, annual 2018-2025, partial 2026, recent June-September and 2019 onward. Each comparison uses the same inclusive opening-timestamp bounds as that BTCUSD window. Slice Binance's native grid by those bounds; do not intersect grids or change the BTCUSD reference accounts. Record each source's actual first/last timestamps, row count, missing four-hour intervals, endpoint coverage and any differing observation counts. A missing scheduled fill expires, and interest spans actual elapsed time, as already registered.

## Scenarios and quote units

Two sources x fourteen windows x ten models x two costs x two execution delays x three borrowing APRs = 3,360 account cases. The unchanged costs are 6/30 bps per dollar or quote unit traded; delays one/three four-hour bars; borrowing APRs 5/10/20%. Initial equity is 10,000 native quote units: USD for the reference inputs, USDT for Binance. Display the quote currency explicitly and use neutral quote-unit field names for monetary results. No USD/USDT conversion, stablecoin basis hedge or FX costs are modeled. This changes venue and quote instrument simultaneously; it cannot isolate pure venue effects or demonstrate equivalent USD investor returns.

Same daily sizing, 0.25-1.50 exposure range, no-borrow sibling, fixed controls, fees, hypothetical 25% maintenance requirement, adverse low-based liquidation, insolvency retention and zero cash yield. No real borrowing terms or account access is implied. Primary comparison remains 10% APR, 6 bps and four-hour delay.

## Reporting and verification

Preserve all outcomes and feature/account traces. Reproduce exactly all 1,560 overlapping BTCUSD account curves from the earlier sizing matrix (thirteen windows x 120 assumptions); the 2018-onward window is new. Verify the previous report and trace hashes before identity comparisons. Require all 23,600 canonical decisions to match and retain future-input/scale account-prefix checks on the new source. Test archive extraction, quarantine identity, checksum rejection, native-grid slicing, quote labels, previous trace comparison and source isolation.

For all six volatility candidates report wins against each source's own BTC hold and simultaneously against BTC, fixed 1.25x and fixed 1.50x, across the twelve assumptions in four core windows: 2018 onward, 2019 onward, 2023 onward and recent. Include annual failures and worst drawdowns/forced sales. Do not use the counts as independent trials, p-values, a parameter selection rule or automatic promotion. If the 60-day lead survives, independent execution realism and genuinely prospective paper evidence remain necessary. If it fails, retain the failed cases and revise the article's interpretation. No production, cloud or current forward-registry changes.